Fig. 10: The comparison between two activation functions.
From: Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

The normalized ϕ(x) is close to tanh(x) at the numerical value and first derivative.
From: Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

The normalized ϕ(x) is close to tanh(x) at the numerical value and first derivative.